Matthew Shaw Undergraduate Dissertation 2014/15

Autonomous Cloud Management

Supervised by M.Stannett

Abstract

Cloud platform providers regularly offer a wide range of generic services, which can be combined with one another (and with user-specific apps) to create very complex composite systems. The number of clients, services and applications available on typical platforms is growing rapidly, and this has obvious management implications. The number of services between which interactions take place means that rogue or faulty services need to be identified quickly; at the same time the proliferation of knock-on effects and emergent behaviours mean that human intervention is no longer enough. Cloud platforms need to be self-managing - services need to monitored in real time to spot problems, and remedial action needs to be taken as quickly as possible to ensure that service level agreements are not violated.

Research has recently begun into techniques for monitoring cloud service characteristics. At the same time, methods based on stream reasoning, the semantic web and cloud sensor networks are being developed which could feasibly allow platforms to replace faulty services with best-possible replacements as and when the need arises. Goals will be to: review research into monitoring of cloud service characteristics, identify areas where further work is required, suggest possible solutions, develop a proof of concept demo system showcasing identified behaviours, present experimental results showing the extent to which solutions behaved as expected